Merge pull request #1167 in HDFFV/hdf5 from ~JHENDERSON/hdf5:bugfix/MPI2_only_fixes to develop

* commit '9aa2eaeb9ed9c94f63973cb55f13de256558497c':
  Add Autotools and CMake checks for big I/O MPI-3 functions
  Add hdf5settings section for parallel compression status in CMake builds
  Switch to CheckSymbolExists in CMake
  Add configure check for MPI_Mprobe and MPI_Imrecv functions

5 files changed, 79 insertions, 22 deletions

diff --git a/src/H5Dio.c b/src/H5Dio.c
index 452105e..5fea91f 100644
--- a/src/H5Dio.c
+++ b/src/H5Dio.c
@@ -1189,7 +1189,8 @@ H5D__ioinfo_adjust(H5D_io_info_t *io_info, const H5D_t *dset,
                                                                     "data transforms needed to be applied",
                                                                     "optimized MPI types flag wasn't set",
                                                                     "one of the dataspaces was neither simple nor scalar",
-                                                                    "dataset was not contiguous or chunked" };
+                                                                    "dataset was not contiguous or chunked",
+                                                                    "parallel writes to filtered datasets are disabled" };
 
                 if(H5CX_get_mpio_local_no_coll_cause(&local_no_collective_cause) < 0)
                     HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "unable to get local no collective cause value")
diff --git a/src/H5Dmpio.c b/src/H5Dmpio.c
index d721ae6..5ebdc06 100644
--- a/src/H5Dmpio.c
+++ b/src/H5Dmpio.c
@@ -229,9 +229,11 @@ static herr_t H5D__mpio_get_sum_chunk(const H5D_io_info_t *io_info,
 static herr_t H5D__construct_filtered_io_info_list(const H5D_io_info_t *io_info,
     const H5D_type_info_t *type_info, const H5D_chunk_map_t *fm,
     H5D_filtered_collective_io_info_t **chunk_list, size_t *num_entries);
+#if MPI_VERSION >= 3
 static herr_t H5D__chunk_redistribute_shared_chunks(const H5D_io_info_t *io_info,
     const H5D_type_info_t *type_info, const H5D_chunk_map_t *fm,
     H5D_filtered_collective_io_info_t *local_chunk_array, size_t *local_chunk_array_num_entries);
+#endif
 static herr_t H5D__mpio_array_gatherv(void *local_array, size_t local_array_num_entries,
     size_t array_entry_size, void **gathered_array, size_t *gathered_array_num_entries,
     hbool_t allgather, int root, MPI_Comm comm, int (*sort_func)(const void *, const void *));
@@ -244,8 +246,10 @@ static herr_t H5D__filtered_collective_chunk_entry_io(H5D_filtered_collective_io
 static int H5D__cmp_chunk_addr(const void *chunk_addr_info1, const void *chunk_addr_info2);
 static int H5D__cmp_filtered_collective_io_info_entry(const void *filtered_collective_io_info_entry1,
     const void *filtered_collective_io_info_entry2);
+#if MPI_VERSION >= 3
 static int H5D__cmp_filtered_collective_io_info_entry_owner(const void *filtered_collective_io_info_entry1,
     const void *filtered_collective_io_info_entry2);
+#endif
 
 
 /*********************/
@@ -330,6 +334,18 @@ H5D__mpio_opt_possible(const H5D_io_info_t *io_info, const H5S_t *file_space,
      *  use collective IO will defer until each chunk IO is reached.
      */
 
+#if MPI_VERSION < 3
+    /*
+     * Don't allow parallel writes to filtered datasets if the MPI version
+     * is less than 3. The functions needed (MPI_Mprobe and MPI_Imrecv) will
+     * not be available.
+     */
+    if (io_info->op_type == H5D_IO_OP_WRITE &&
+        io_info->dset->shared->layout.type == H5D_CHUNKED &&
+        io_info->dset->shared->dcpl_cache.pline.nused > 0)
+        local_cause |= H5D_MPIO_PARALLEL_FILTERED_WRITES_DISABLED;
+#endif
+
     /* Check for independent I/O */
     if(local_cause & H5D_MPIO_SET_INDEPENDENT)
         global_cause = local_cause;
@@ -2127,6 +2143,7 @@ H5D__cmp_filtered_collective_io_info_entry(const void *filtered_collective_io_in
     FUNC_LEAVE_NOAPI(H5F_addr_cmp(addr1, addr2))
 } /* end H5D__cmp_filtered_collective_io_info_entry() */
 
+#if MPI_VERSION >= 3
 
 /*-------------------------------------------------------------------------
  * Function:    H5D__cmp_filtered_collective_io_info_entry_owner
@@ -2157,6 +2174,7 @@ H5D__cmp_filtered_collective_io_info_entry_owner(const void *filtered_collective
 
     FUNC_LEAVE_NOAPI(owner1 - owner2)
 } /* end H5D__cmp_filtered_collective_io_info_entry_owner() */
+#endif
 
 
 /*-------------------------------------------------------------------------
@@ -2585,8 +2603,12 @@ H5D__construct_filtered_io_info_list(const H5D_io_info_t *io_info, const H5D_typ
 
     /* Redistribute shared chunks to new owners as necessary */
     if (io_info->op_type == H5D_IO_OP_WRITE)
+#if MPI_VERSION >= 3
         if (H5D__chunk_redistribute_shared_chunks(io_info, type_info, fm, local_info_array, &num_chunks_selected) < 0)
             HGOTO_ERROR(H5E_DATASET, H5E_WRITEERROR, FAIL, "unable to redistribute shared chunks")
+#else
+        HGOTO_ERROR(H5E_DATASET, H5E_WRITEERROR, FAIL, "unable to redistribute shared chunks - MPI version < 3 (MPI_Mprobe and MPI_Imrecv missing)")
+#endif
 
     *chunk_list = local_info_array;
     *num_entries = num_chunks_selected;
@@ -2595,6 +2617,7 @@ done:
     FUNC_LEAVE_NOAPI(ret_value)
 } /* end H5D__construct_filtered_io_info_list() */
 
+#if MPI_VERSION >= 3
 
 /*-------------------------------------------------------------------------
  * Function:    H5D__chunk_redistribute_shared_chunks
@@ -2893,6 +2916,7 @@ done:
 
     FUNC_LEAVE_NOAPI(ret_value)
 } /* end H5D__chunk_redistribute_shared_chunks() */
+#endif
 
 
 /*-------------------------------------------------------------------------
diff --git a/src/H5Ppublic.h b/src/H5Ppublic.h
index 58f3d18..c5596e5 100644
--- a/src/H5Ppublic.h
+++ b/src/H5Ppublic.h
@@ -166,7 +166,8 @@ typedef enum H5D_mpio_no_collective_cause_t {
     H5D_MPIO_MPI_OPT_TYPES_ENV_VAR_DISABLED = 0x08,
     H5D_MPIO_NOT_SIMPLE_OR_SCALAR_DATASPACES = 0x10,
     H5D_MPIO_NOT_CONTIGUOUS_OR_CHUNKED_DATASET = 0x20,
-    H5D_MPIO_NO_COLLECTIVE_MAX_CAUSE = 0x40
+    H5D_MPIO_PARALLEL_FILTERED_WRITES_DISABLED = 0x40,
+    H5D_MPIO_NO_COLLECTIVE_MAX_CAUSE = 0x80
 } H5D_mpio_no_collective_cause_t;
 
 /********************/
diff --git a/src/H5Smpio.c b/src/H5Smpio.c
index db81ffc..2bd275a 100644
--- a/src/H5Smpio.c
+++ b/src/H5Smpio.c
@@ -271,29 +271,58 @@ H5S_mpio_create_point_datatype (size_t elmt_size, hsize_t num_points,
       if(NULL == (inner_disps = (MPI_Aint *)H5MM_malloc(sizeof(MPI_Aint) * (size_t)total_types)))
           HGOTO_ERROR(H5E_DATASPACE, H5E_CANTALLOC, FAIL, "can't allocate array of blocks")
 
+#if MPI_VERSION < 3
+      /* Allocate block sizes for MPI datatype call */
+      if(NULL == (blocks = (int *)H5MM_malloc(sizeof(int) * bigio_count)))
+          HGOTO_ERROR(H5E_DATASPACE, H5E_CANTALLOC, FAIL, "can't allocate array of blocks")
+
+      for(u = 0; u < bigio_count; u++)
+          blocks[u] = 1;
+#endif
+
       for(i=0 ; i<num_big_types ; i++) {
+#if MPI_VERSION >= 3
           if(MPI_SUCCESS != (mpi_code = MPI_Type_create_hindexed_block(bigio_count,
                                                                        1,
                                                                        &disp[i*bigio_count],
                                                                        elmt_type,
                                                                        &inner_types[i]))) {
-             HMPI_GOTO_ERROR(FAIL, "MPI_Type_create_hindexed_block failed", mpi_code);
+              HMPI_GOTO_ERROR(FAIL, "MPI_Type_create_hindexed_block failed", mpi_code);
           }
+#else
+          if(MPI_SUCCESS != (mpi_code = MPI_Type_create_hindexed((int)bigio_count,
+                                                                      blocks,
+                                                                      &disp[i*bigio_count],
+                                                                      elmt_type,
+                                                                      &inner_types[i]))) {
+              HMPI_GOTO_ERROR(FAIL, "MPI_Type_create_hindexed failed", mpi_code)
+          }
+#endif
           inner_blocks[i] = 1;
           inner_disps[i]  = 0;
       }
 
       if(remaining_points) {
+#if MPI_VERSION >= 3
           if(MPI_SUCCESS != (mpi_code = MPI_Type_create_hindexed_block(remaining_points,
                                                                        1,
                                                                        &disp[num_big_types*bigio_count],
                                                                        elmt_type,
                                                                        &inner_types[num_big_types]))) {
               HMPI_GOTO_ERROR(FAIL, "MPI_Type_create_hindexed_block failed", mpi_code);
-	  }
+	      }
+#else
+          if(MPI_SUCCESS != (mpi_code = MPI_Type_create_hindexed((int)remaining_points,
+                                                                      blocks,
+                                                                      &disp[num_big_types*bigio_count],
+                                                                      elmt_type,
+                                                                      &inner_types[num_big_types]))) {
+              HMPI_GOTO_ERROR(FAIL, "MPI_Type_create_hindexed failed", mpi_code)
+          }
+#endif
           inner_blocks[num_big_types] = 1;
           inner_disps[num_big_types] = 0;
-        }
+      }
 
       if(MPI_SUCCESS != (mpi_code = MPI_Type_create_struct(total_types,
                                                            inner_blocks,
diff --git a/src/libhdf5.settings.in b/src/libhdf5.settings.in
index 51b24dc..531cd00 100644
--- a/src/libhdf5.settings.in
+++ b/src/libhdf5.settings.in
@@ -67,20 +67,22 @@ Languages:
 
 Features:
 ---------
-                  Parallel HDF5: @PARALLEL@
-             High-level library: @HDF5_HL@
-                   Threadsafety: @THREADSAFE@
-            Default API mapping: @DEFAULT_API_VERSION@
- With deprecated public symbols: @DEPRECATED_SYMBOLS@
-         I/O filters (external): @EXTERNAL_FILTERS@
-                            MPE: @MPE@
-                     Direct VFD: @DIRECT_VFD@
-                        dmalloc: @HAVE_DMALLOC@
- Packages w/ extra debug output: @INTERNAL_DEBUG_OUTPUT@
-                    API tracing: @TRACE_API@
-           Using memory checker: @USINGMEMCHECKER@
-Memory allocation sanity checks: @MEMORYALLOCSANITYCHECK@
-            Metadata trace file: @METADATATRACEFILE@
-         Function stack tracing: @CODESTACK@
-      Strict file format checks: @STRICT_FORMAT_CHECKS@
-   Optimization instrumentation: @INSTRUMENT_LIBRARY@
+                   Parallel HDF5: @PARALLEL@
+Parallel Filtered Dataset Writes: @PARALLEL_FILTERED_WRITES@
+              Large Parallel I/O: @LARGE_PARALLEL_IO@
+              High-level library: @HDF5_HL@
+                    Threadsafety: @THREADSAFE@
+             Default API mapping: @DEFAULT_API_VERSION@
+  With deprecated public symbols: @DEPRECATED_SYMBOLS@
+          I/O filters (external): @EXTERNAL_FILTERS@
+                             MPE: @MPE@
+                      Direct VFD: @DIRECT_VFD@
+                         dmalloc: @HAVE_DMALLOC@
+  Packages w/ extra debug output: @INTERNAL_DEBUG_OUTPUT@
+                     API tracing: @TRACE_API@
+            Using memory checker: @USINGMEMCHECKER@
+ Memory allocation sanity checks: @MEMORYALLOCSANITYCHECK@
+             Metadata trace file: @METADATATRACEFILE@
+          Function stack tracing: @CODESTACK@
+       Strict file format checks: @STRICT_FORMAT_CHECKS@
+    Optimization instrumentation: @INSTRUMENT_LIBRARY@